Julian, thank you for the thorough explanation of the issues of this, this helps in understanding the magnitude of this task. I think one of the issues may be in difference between what I will call a “Play Out Server” and a “Production Media Server”
So with a Play Out Server, they are typically remotely managed, schedules loaded for a week, sending video into a cable system or TV station switcher, they switch to the server when they need to. The Play Out Server will just play file after file for a 24 hour schedule, no layers, no graphics- just one file ends, and the next file plays. Some have GPI triggers that will start playback when someone in the control room triggers it, they switch to it, put its content on air and switch back to other sources when done but 90% of these server play for days/ weeks without anyone ever touching the box. Sometimes this is called - channel in a box, not glamorous or very fun but these work great for unmanned channels, overnights and paid show placement.
With a Production Media Server, you have all the fun stuff, layers, graphics, keying, ext. Basically CasparCG, there are some high end Production servers that do all of these things while maintaining captions and also able to just play 24/7. I am not totally sure on how they maintain captions with layers / graphics and all but pretty sure there is just a base layer with the main video/ captions that will pass, and stated in the manual that if you want captions to pass, they need to be on this layer where they are sent directly to the output and combined with the final output video. Never heard of captions being combined.
From what I have read, I believe you are right about the Decklink needing 10 bit YUV frames for captions and I see the need for an alpha channel in a Production server but not in the Play Out model.
So with your info, if this was to be done it seems we would need a Play Out server version that simply sends “local files” directly to the “consumer”/ Decklink as viewed in the CasparCG_Server.jpg This would be a stripped down version with no Producers, Mixers, layers or Keying. But then there is still the problem of getting YUV frames. It may be that just making a player in FFMPEG would be better for a simple Play Out server but have had problems even getting FFMPEG to pass captions. Wish it would have been easier to do in CasparCG but at least now the complications are known. Please let me know if there are any other solutions for getting captions to pass out Decklink SDI and thanks again!